An approach of soft-computing in optimizing controlled release products

This paper presents a solution for optimizing controlled release product formulation using a combination of AI techniques (Soft-Computing): neural networks, fuzzy logic and genetic algorithms. This achievement will help to significantly reduce time and labour in R&D process thank to its good accuracy and high processing speed. The results obtained from this research indicate that the alternative approach can be considered as an effective and efficient method for modelling and optimising controlled release formulations.

TẠP CHÍ PHÁT TRIỂN KH&CN, TẬP 13, SỐ T2 - 2010 AN APPROACH OF SOFT-COMPUTING IN OPTIMIZING CONTROLLED RELEASE PRODUCTS Nam Phuong Nguyen, Nam Huu Bui, Duong Quang Do University of Medicine and Pharmacy Ho Chi Minh City ABSTRACT: In the pharmaceutical market, all products have a life cycle Out of date products should be replaced by new ones, which have better quality For this reason, modelling and optimizing formulation are the regular demands Traditional methods of design and optimization - such as statistics, simplex – can only be used for simple and linear data In case of complicated or non-linear data, alternative methods that are able to deal with such data are needed This paper presents a solution for optimizing controlled release product formulation using a combination of AI techniques (Soft-Computing): neural networks, fuzzy logic and genetic algorithms This achievement will help to significantly reduce time and labour in R&D process thank to its good accuracy and high processing speed The results obtained from this research indicate that the alternative approach can be considered as an effective and efficient method for modelling and optimising controlled release formulations Keywords: Neural networks, Genetic Algorithms, Optimization, Soft computing, Controlled Release space to find the point, which has the optimum INTRODUCTION balance of properties Nowadays, formulators Formulation design is regular work of pharmacist because all products have a life cycle products quality need to be constantly improved Out of date products should be replaced by better ones For this reason, modelling and optimization of formulation are the regular demands [1] Traditional methods of design and optimization of formulation - such as statistics, simplex, - are only used for simple and linear data These methods are not suitable for complicated or non-linear data The formidable task of formulation research is to navigate multidimensional design can develop complex dosage forms by design and optimization way Since product properties are affected not only by the ratio in which the ingredients are combined but also by the processing parameters, the ingredient levels and processing conditions should be taken into account in formulation design Computer technology in the form of artificial intelligence provides an affordable means of improvement in product formulation and has more promising of solving an optimization of product formulation because it is not finite of ingredients (X) and can simultaneously Trang 71 Science & Technology Development, Vol 13, No.T2- 2010 optimize many properties (Y) of the effect, particularly to nonlinear and complex formulation and is suitable for the problems data A comparison can be observed for ANN with complicated and non-linear data with mammalian nerve connectivity The In this study, a combination of neural mammalian nervous system is built up from networks, fuzzy logic and genetic algorithms biological neurons Each neuron collects input (GA) called Soft-Computing (SC) is employed stimuli and triggers an output to the next with neural networks considered as a method neurons in the system (see Figure 1) Similarly, for modelling whilst GA combined with fuzzy artificial logic acted to optimisation process Each of connecting signal and nodes that collect techniques has advantages and disadvantages, mathematical inputs and produce the output but if they are accurately combined all signals that are passed to the next neurons [6, 7] together, the disadvantages of this will be [2, 3, 4, 5, 6] neural networks also involve The units in the input layer only have one - input signal assigned to them, while the nodes for example, neural networks is difficult to in the hidden layer are connected and assigned extract knowledge, but fuzzy inference systems by many of the input signals The output layer does it easily The paper then reports the depends upon the structure of network in that application of SC to two sets of published there are only one or many output nodes with formulation data, one for a matrix tablet, and respectively many or a unique output signal the other for controlled release microspheres An artificial neural network is generally and compares the results obtained with composed of several layers: input layer, hidden statistical analyses layers (one or many), and output layer For overcome by advantages of another example, the structure of a neural network with SOFT-COMPUTING CONCEPT inputs, output, and nodes in a single 2.1 Modelling formulation data with Neural hidden layer is detailed in Figure However, networks neural networks are often known as “black Neural networks are complementary box” technologies in that the means of mapping adaptive inputs to output(s) is hidden within the network intelligent systems Artificial Neural Network structure It is also quite different from (ANN) learns from scratch by adjusting the statistical methods in that a neural network interconnections between layers For over 60 does not produce a mathematical equation years, ANNs have been applied to design a Neural networks are often used to design model of relationships between cause and predictive models technologies Trang 72 in the design of TẠP CHÍ PHÁT TRIỂN KH&CN, TẬP 13, SỐ T2 - 2010 Synapse Dendrites Axon Cell body Figure Structure of biological neuron input input output input output input Input layer Output layer Hidden layer Figure Structure of neural networks 2.2 The combination model of GA and fuzzy evolutionary processes (detailed in Figure 3) logic for optimization This step, genetic algorithms associate with fuzzy logic to optimise formulation - the fitness Genetic Algorithms (GA) are derivative- function free stochastic optimization methods based on natural selection relationships and Initialize a population Start Generate a new population Evaluate Fitness value Meet Stop conditions? Yes Select Best formulation cause-and-effect No Mutation of Crossover concepts on [8] Reproduction the based End Figure The cycle of selection and evolutionary processes with GA Trang 73 Science & Technology Development, Vol 13, No.T2- 2010 Given a way or a method of encoding solution of a problem into the form of recombination as the “parent’ formulations mate chromosomes and given an evaluation function that returns a measurement of the cost value of Step 4: Delete members of the population to make room for new formulations any chromosome in the context of the problem, the processing GA includes steps [6, 9] Step 5: Evaluate the new formulations and insert them into the population Step 1: Initialize a set of solutions (potential formulations) randomly - called population Step 6: If the stopping criterion is satisfied, then stop and return the optimum formulations; otherwise, go to Step Step 2: Evaluate each formulation in the population The detailed membership functions from the fuzzy logic, applied to optimization with Step 3: Create new formulations by mating GA, are as follows: current formulations; apply mutation and Flat-Tent function desirability drops between Mid1 (a): Flat function (d): any value is linearly and acceptable; the the maximum, acceptable are That values is but between Mid1 and Mid2, the values membership function in the set of acceptable minimum, and between Mid2 and its (a) perfectly is, (d) their membership function in the set of acceptable values is Up-Hill function (b): any value Down-Hill function (c): any value between the mid-point (Mid1 between the mid-point (Mid1 = = Mid2) and the maximum is Mid2) completely acceptable; its and the minimum is completely membership function in the set acceptable; its from mid-point to of acceptable values is Any value from minimum to midpoint, the desirability decreases linearly until it is zero at the minimum point Trang 74 maximum, (b) the desirability membership function in the set of acceptable values is Any value decreases linearly until it is zero at the maximum point (c) TẠP CHÍ PHÁT TRIỂN KH&CN, TẬP 13, SỐ T2 - 2010 A fusion of neural networks, fuzzy logic 2.3 Solving the optimization problem with neural networks, fuzzy logic and genetic and genetic algorithms to deal with an algorithms optimization of product formulation problem is illustrated in Figure Neural networks Formulation and product variables Optimal requirements No Genetic algorithms & Fuzzy logic Cause-and-effect model Optimization & Evaluation Yes Optimized formulation Figure The cycle of modelling and optimization The detailed processing of optimization is 2.4.Software tool as follows: Step The software was used in this research is 1: establish cause-and-effect BCPharSoft OPT This is a software tool, relationship by using neuro-fuzzy system or which neural networks language It was a modified form of that Step 2: determine optimal requirements defined by user Step 3: optimize ingredients corresponding to optimal condition of properties by using genetic algorithms combined to fuzzy logic, the is built described in C#.net previously programming – INForm (www.intelligensys.co.uk), but with additional functionalities in order to improve the quality of predictive models and the optimum formulation fitness function of GA is cause-and-effect In order to evaluate the quality of a relationship model determined from Step predictive model generated by ANN, the Repeat Step until a stopping criterion is met correlation coefficient R-squared (R2) was or optimal condition is reached computed, with higher values of R2 indicating the improved quality of the model [10] Trang 75 Science & Technology Development, Vol 13, No.T2- 2010 R    = 1 −     n ∑ ( y i − yˆ i )  i =1 n  x100  ( y y ) − ∑ i  i =1  where y : the mean of the dependent variable; yˆ : the predicted value from the model; n: number of records were prepared using sodium alginate as a EXPERIMENTAL DATA polymer and CaCl2 as a cross-linking agent A The formulation database of the matrix tablet taken from the literature (Bodea and Leucuta, 1997) [11] , consisted of 14 experimental records, and involved varying percentages of two hydrophilic polymers (hydroxypropylmethylcellulose, HPMC - X1, sodium carboxymethylcellulose, CMCNa - X2) and propranolol HCL - X3 The measured outputs were the cumulative percentages of drug released after 1, 6, and 12h sampling intervals (Y1, Y2, and Y3, respectively) These data were modelled and optimised in the original study [11] by statistical methods using a D-optimal quadratic model In the present 33 full factorial design was used to investigate the joint influences of three variables - the stirring speed during preparation of the microspheres (X1), concentration of CaCl2 (X2) and % of heavy liquid paraffin in a blend of heavy and light liquid paraffin in the dispersion medium (X3) - on the time for 80% drug dissolution (t80) In addition, in the published study [12] , the % drug released after 60 (Y60), 360 (Y360), and 480 (Y480) was also considered as outputs that were analysed 25 records were used as training data, and records used as unseen data to test predictive power study, 11 records were used for training and records used as unseen data for testing the predictive models Another formulation EXPERIMENTAL RESULTS 4.1 Matrix tablet formulation database for controlled release diclofenac sodium microspheres containing 27 experimental records taken from a published paper (Gohel and Amin, 1998) [12] was used for validating the capability of SC for such of formulation as well In this study, microspheres Trang 76 By selecting suitable values of control parameters, SC generated satisfactory models for all responses of the matrix tablet formulation The correlation coefficient R2 values of the predictive models generated from SC were showed in Table TẠP CHÍ PHÁT TRIỂN KH&CN, TẬP 13, SỐ T2 - 2010 Table R2 values of the predictive models generated from SC and statistical method [11] Method Y1 Y2 Soft-Computing R Train= 0.98 R Train = 0.99 R Train= 0.99 Statistical Y3 R Test= 0.99 R Test= 0.9 R2 Test = 0.99 R2 = 0.98 R2 = 0.97 R2 = 0.99 R2 = 0.96 R2 = 0.88 R2 = 0.91 , the for the models of the cumulative percentage present study gave improved models for all release after 1h (Y1), 6h (Y2) and 12h (Y3), the responses The analyses in Table showed that quality of the models was improved with Compared with a published study [11] significantly higher R2 values Predicted 0.14 0.12 yâ -SC = 0.90x + 0.012 yâ -Stat = 0.96x + 0.004 yâ -SC = 1.07x - 0.037 0.7 yâ -Stat = 0.89x + 0.063 R2 = 0.90 R2 = 0.96 0.1 0.08 0.6 0.5 0.4 0.06 0.04 0.04 0.8 R2 = 0.98 R2 = 0.99 Predicted 0.16 y1-SC y1-Stat y2-SC y2-Stat 0.3 0.06 0.08 0.1 Observed 0.12 0.14 0.16 0.3 0.4 0.5 0.6 Observed 0.7 0.8 yâ -SC = 0.98x + 0.018 1.1 R2 = 0.99 yâ -Stat = 0.92x + 0.068 Predicted R2 = 0.92 0.9 0.8 0.7 y3-SC y3-Stat 0.6 0.6 0.7 0.8 0.9 1.1 Observed Figure Scatter plots, linear equations and R2 values for the observed data points from SC and statistical methods for Y1, Y2 and Y3 In comparison with the statistical result from the SC models were much improved reported in the literature [11] showed in Figure 5, compared to those from the statistical models the linear R2 values for all observed responses All of these results proved that overall the were significantly higher to those from the predictive models generated from SC were statistical models Moreover, for the outputs Y2 superior and Y3, the slope and the intercept coefficients when compared presented in the literature [11] to the results Trang 77 Science & Technology Development, Vol 13, No.T2- 2010 For the optimisation of this product, the As showed in Table 2, SC generated constraints of optimum formulation used in this several optimum formulations for this product study were also taken from the literature that that met all optimum conditions mentioned was as follows: above In addition, when compared to a single [11] X2+X3 ≤ 0.8 0.1 ≤ Y1 ≤ 0.2 outcome optimised from statistical method X3 ≥ 0.34 0.45 ≤ Y2 ≤ 0.55 this approach is definitely superior because of 0.8 ≤ Y3 its multiple formulations optimised , Table Optimum formulations generated from SC X1 X2 X3 Y1 Y2 Y3 (1) 0.453 0.007 0.519 0.145 0.546 0.843 (2) 0.334 0.101 0.508 0.113 0.550 0.907 (3) 0.316 0.101 0.498 0.104 0.548 0.908 From Table 2, it also demonstrated that though SC generated different optimum 4.2 Controlled release diclofenac sodium microspheres formulation formulations generated, they still met the It is similar to the first data, by selecting required constraint The first formulation suitable values of control parameters, the showed the maximum value for Y1, the second correlation coefficient R2 values of the formulation showed the maximum value for predictive models for the diclofenac sodium Y2, while Y3 obtained the maximum value with microspheres formulation generated from SC the third formulation For these formulations were showed in Table The results in Table the formulators could get more selections for showed that for this product SC achieved their different purposes, for example if they significantly higher quality predictive models want to maximize the % of drug dissolved in for all responses In particular, SC predicted a 6h (Y2) and optimize the formulation of this model with R2 = 0.93 for Y60 whilst statistical drug following the constrains showed above, method gave R2 = 0.74 only for this property they could consider the second formulation as the optimum one by themselves Table R2 values of the predictive models generated from SC and statistical method [12] Method Soft-Computing Trang 78 t80 Y60 Y360 2 Y480 R Train= 0.99 R Train= 0.93 R Train= 0.97 R2 Train= 0.97 R2 Test= 0.95 R2 Test= 0.84 R2 Test= 0.99 R2 Test= 0.96 TẠP CHÍ PHÁT TRIỂN KH&CN, TẬP 13, SỐ T2 - 2010 R2 = 0.93 R2 = 0.98 R2 = 0.97 0.99 0.74 0.95 0.92 600 yâ t80 -SC = 0.95x + 18.89 R2 = 0.99 yâ t80 -Stat = 0.96x + 15.89 Predicted 500 R2 = 0.96 400 300 t80-SC 200 200 55 Predicted 300 400 Observed t80-Stat 500 600 yâ60-SC = 0.96x + 2.07 R = 0.94 50 yâ 60 -Stat = x - 4.51 45 R = 0.49 40 35 30 y60-SC y60-Stat 25 25 30 35 40 45 50 55 Predicted Observed 90 yâ360-SC = 0.95x + 3.93 85 R = 0.98 yâ360-Stat = 0.73x + 22.30 80 R = 0.81 2 75 70 65 y360-SC 60 60 100 65 70 75 Observed 80 y360-Stat 85 90 yâ 480 -SC = 0.98x + 1.97 R2 = 0.98 95 Predicted Statistical R2 = 0.99 yâ 480 -Stat = 0.92x + 6.95 90 R2 = 0.92 85 80 75 y480-SC y480-Stat 70 70 75 80 85 Observed 90 95 100 Figure Scatter plots, linear equations and R2 values for the observed data points from SC and statistical methods for t80, Y60, Y360 and Y480 Trang 79 Science & Technology Development, Vol 13, No.T2- 2010 From Figure 6, it is clear that the For the optimisation of this product, the satisfactory predictive power of the SC models constraints of optimum formulation used in this for the observed data can be seen The linear R study were also taken from the literature that values for all these responses were significantly was as follows: 20% ≤ Y60 ≤ 40%, 50% ≤ Y360 high and the slope and the intercept coefficients ≤ 70% 65% ≤ Y480 ≤ 80% and X1: integer from the SC models were acceptable as well In As showed in Table 4, SC generated several general in comparison with the statistical optimum formulations for this product that met method, SC produced satisfactory models for all optimum conditions mentioned above In all responses Moreover for the Y60 response, addition, when compared to a single outcome the predictive model of SC for this formulation optimised from statistical method significantly overcame the result generated approach is definitely superior because of its from statistical analysis multiple formulations optimised [12] [12] , this Table Optimum formulations generated from SC X1 X2 X3 t80 Y60 Y360 Y480 (1) 500 14.13 32.79 560.31 39.99 67.77 77.17 (2) 1500 14.31 49.30 470.98 39.03 69.90 79.00 (3) 540 7.50 44.75 482.26 37.69 69.40 80.00 From Table 4, it also demonstrated that output is fixed and if a formulator wants to though SC generated completely different improve the quality of the final statistical optimum formulations generated, they also met equation, he must carry out further experiments the required constraint For these formulations to obtain a higher quality data set However the formulators could get more selections for with SC, a formulator can obtain alternative their own purposes outputs, with a selection of an appropriate 4.3 General comments training model For example, by changing values of control parameters, the quality of the When validating the capability of SC and predictive equation can be improved In other comparing the predictive power of this method words, a formulator can perform SC in an to the statistical methods for both controlled iterative manner by directed change of control release products, it was recognised that the parameter values until the most appropriate basis of the statistical approach is to use and/or predictive model is obtained Moreover, standard equations and procedures based on a single optimised formulation generated from statistical theory to obtain the final equation statistical considered as predictive model The statistical Trang 80 analysis is also a major TẠP CHÍ PHÁT TRIỂN KH&CN, TẬP 13, SỐ T2 - 2010 inconvenience of this method when compared solving an optimization of product formulation to SC in pharmaceuticals is an example This solution helps formulator reduce time and labor more CONCLUSIONS than traditional methods In contrast to Although neural networks, fuzzy logic and statistical approaches, Soft-computing, with its genetic algorithms had been introduced for a advantage of generating several optimum long time, applications using theories of neural formulations and superior predictive models, networks, fuzzy logic and genetic algorithm are has been shown to be an efficient method for still interested; the application using the neural modelling and optimising controlled release networks, fuzzy logic and genetic algorithm for formulations ỨNG DỤNG KỸ THUẬT TÍNH TỐN MỀM GIẢI QUYẾT BÀI TỐN TỐI ƯU HĨA CƠNG THỨC VIÊN PHĨNG THÍCH CĨ KIỂM SỐT Nguyễn Phương Nam, Bùi Hữu Nam, Đỗ Quang Dương Trường Đại học Y Dược Tp Hồ Chí Minh TĨM TẮT: Trong ngành dược, nhà sản xuất thuốc việc thành lập tối ưu hóa cơng thức việc làm thường xuyên sản phẩm có vòng đời định nhu cầu cạnh tranh thị trường đòi hỏi phải khơng ngừng cải tiến sản phẩm có hay thay sản phẩm Chính lý này, tối ưu hóa cơng thức dược đề cập đến Các phương pháp tối ưu hóa truyền thống (tốn thống kê, đơn hình…) áp dụng với liệu đơn giản tuyến tính Chúng khơng phù hợp với liệu phức tạp phi tuyến Ngoài ra, phương pháp truyền thống khơng tối ưu hóa đồng thời nhiều biến phụ thuộc sản phẩm thường có nhiều tính chất Phương pháp tối ưu hóa thơng minh có nhiều triển vọng thay phương pháp truyền thống Bài báo đưa phương pháp tối ưu hóa thơng minh Đó kết hợp mạng thần kinh, logic mờ thuật toán di truyền Phương pháp giải khó khăn mà phương pháp truyền thống thực Các kết thu từ nghiên cứu chứng minh phương pháp tối ưu hóa hiệu Từ khóa: mạng thần kinh, logic mờ, thuật tốn di truyền, kỹ thuật tính tốn mềm Trang 81 Science & Technology Development, Vol 13, No.T2- 2010 Architecture for Adaptive Learning and REFERENCES [1] Giap D.V Intelligence Applying Software Knowledge Acquisition Department of an for Artifical Optimizing Formulation and Process Faculty of Pharmacy - University of Medicine and INForm 3.0ation Science,University of Otago, P.O.Box 56, Dunedin, New Deepa S.N Zealand [8] Sivanandam S.N., Pharmacy at Ho Chi Minh City, Viet Nam Introduction to Genetic Algorithms ISBN (2001) 978-3-540-73189-4 Berlin, Heidelberg, [2] Nauck D., Kruse R Neuro_Fuzzy Systems for Function Approximation Faculty of New York Springer (2008) [9] Kiem H., Thai L.H Genetic Algorithms: Computer Science_ Neural and Fuzzy Solving Systems, naturally Otto_von_Guericke_University of Magdeburg, Universitaetsplatz 2, D- J Neurofuzzy problems Viet Nam, in computer Educational Publisher (2000) [10] Shao Q., Rowe R.C., York P Comparison 39106 Magdeburg, Germany (1997) [3] Jantzen the Modelling of neurofuzzy logic and neural networks in Department of Automation, University of modelling Denmark, report no 98-H-874 (Oct 1998) immediate release tablet formulation Eur [4] Jang J.R., Sun C., Mizutani E Neuro-fuzzy experimental data of an J Pharm Sci, Vol 28, pp 394–404 (2006) and Soft Computing: A Computation [11] Bodea A., Leucuta S.E Optimization of Approach To Learning And Machine hydrophilic matrix tablets using D-optimal Intelligence Upper Saddle River, Prentice- design Int J Pharm, Vol 153, pp 247- Hall (1997) 255 (1997) [5] Bonissone P.P Adaptive Neural Fuzzy [12] Gohel M.C., Amin A.F Formulation Inference Systems (ANFIS): Analysis and optimization Applications GE CRD, Schenectady, NY diclofenac sodium microspheres using USA (1997) factorial design J Control Release, Vol [6] Lin C.T., George Lee C.S Neural Fuzzy Systems: A Neuro-Fuzzy Synergism to Intelligent System International ed, ISBN 0-13-261413-8 Upper Saddle River, Prentice-Hall (1996) [7] Kasabov N.K., Kim J.S., Gray A.R., Watts M.J FuNN - A Fuzzy Neural Network Trang 82 of controlled 51, pp 115-122 (1998) release ... variables - the stirring speed during preparation of the microspheres (X1), concentration of CaCl2 (X2) and % of heavy liquid paraffin in a blend of heavy and light liquid paraffin in the dispersion... of an appropriate 4.3 General comments training model For example, by changing values of control parameters, the quality of the When validating the capability of SC and predictive equation can... Learning and REFERENCES [1] Giap D.V Intelligence Applying Software Knowledge Acquisition Department of an for Artifical Optimizing Formulation and Process Faculty of Pharmacy - University of Medicine